Detección de fallas en cajas de engranajes utilizando el método de aprendizaje de máquinas Support Vector Machine (SVM)

Abstract

The objective of this research was to create a predictive model under the machine learning approach and verify its effectiveness to classify and detect faults in gearboxes automatically, for which a data set of vibration signals obtained from the repository was used from the Open Energy Data Initiative (OEDI) of the US Department of Energy. The creation of the model was carried out using the Support Vector Machine (SVM) supervised machine learning method and with the aid of Python programming software, where the preprocessing and analysis of the data set was performed. Features in the time domain and frequency domain were extracted from the data set. To select the best features, the Recursive Features Elimination with Cross Validation (RFECV) method was applied. To enter the SVM classifier, the data was divided into 70% for training and 30% for testing. As a result, three fault detection models were obtained, a first model where a set of data collected by four accelerometers under a load of 50% was produced, a second model where the data collected by four accelerometers and loads in a range of 0 to 90% and a third model using the data from a single accelerometer of model two. Each model was trained and tested obtaining excellent results, achieving an accuracy of 99,84% and a precision of 99,82% for the best model. The results show that the method used classifies and predicts faults with high accuracy and precision, being a promising method and of great contribution to industrial maintenance. It is recommended to reduce and standardize the set of features, in this way it is possible to reduce the computational load and in turn improve the performance of the model.